diff --git a/model/train_nmt.py b/model/train_nmt.py
index c03f184..a1a4057 100644
--- a/model/train_nmt.py
+++ b/model/train_nmt.py
@@ -3,7 +3,12 @@
 
 from tensorflow.python.keras.utils import to_categorical
 import numpy as np
-import os
+import os, sys
+
+project_path = os.path.sep.join(os.path.abspath(__file__).split(os.path.sep)[:-2])
+if project_path not in sys.path:
+    sys.path.append(project_path)
+
 from utils.data_helper import read_data, sents2sequences
 from model.nmt import define_nmt
 import matplotlib.pyplot as plt
@@ -12,8 +17,8 @@
 def get_data(train_size, random_seed=100):
 
     """ Getting randomly shuffled training / testing data """
-    en_text = read_data(os.path.join('..', 'data', 'small_vocab_en.txt'))
-    fr_text = read_data(os.path.join('..', 'data', 'small_vocab_fr.txt'))
+    en_text = read_data(os.path.join(project_path, 'data', 'small_vocab_en.txt'))
+    fr_text = read_data(os.path.join(project_path, 'data', 'small_vocab_fr.txt'))
     print('Length of text: {}'.format(len(en_text)))
 
     fr_text = ['sos ' + sent[:-1] + 'eos .'  if sent.endswith('.') else 'sos ' + sent + ' eos .' for sent in fr_text]
@@ -184,7 +189,6 @@ def plot_attention_weights(encoder_inputs, attention_weights, en_id2word, fr_id2
     print('Translating: {}'.format(test_en))
 
     test_en_seq = sents2sequences(en_tokenizer, [test_en], pad_length=en_timesteps)
-    print(test_en_seq)
     test_fr, attn_weights = infer_nmt(
         encoder_model=infer_enc_model, decoder_model=infer_dec_model,
         test_en_seq=test_en_seq, en_vsize=en_vsize, fr_vsize=fr_vsize)